Article Preview
Top1. Introduction
IaaS model enables the customers to utilize the VMs as per their resource needs on pay-as-you-use model. This permits clients to run their applications on the most suitable and customized VMs and pay for the exact usage. Nonetheless, the resources provided by cloud suppliers in terms of CPU, memory, disc and network bandwidth always varies dynamically to fulfill dynamic demands of customer for resizing or termination of VMs. As an outcome, if cloud service providers fails to reallocate heterogeneous VMs and resources, a few hosts may become over-burden while other hosts might be underutilized. In the long run, such an uneven utilization of hosts brings about pointless initiation of servers, in this manner expanding the overall expenses.
Also, allocation of most of VMs to some over-burdened servers will potentially influences the QoS and causes SLA infringement. The VM live migration plays a vital role in VM consolidation, subsequently reallocate all the VMs into a couple of physical servers and power-off unutilized servers. This methodology improves the overall utilization of resources and permits energy saving while maintaining QoS.
A live-migration is an expensive activity. It causes additional network bandwidth utilization, CPU utilization and VM downtime. At the point when multiple VMs to be moved, it is imperative to plan the migrations wisely, so as to limit effect on QoS of both source and destination hosts and migrating VM.
VM migration issues are carefully examined, but we see that the VMs scheduling policies don't get a similar degree of consideration. It results in the inaccurate VM scheduling which leads to QoS degradation causing SLA infringement and we can't achieve the goal of energy saving with full accuracy.
This paper introduces POGS: Pareto-Optimal Green Scheduler, a live migration-based scheduler that efficiently calculates the migration overheads and selects the best host to achieve a seamless VM live migration. The assessment of POGS is performed over KVM based VMs as explained further sections.